Microstructure and physical properties of hydrate-bearing sediment: Digital core and deep learning application

Abstract

Revealing the microstructural characteristics of hydrate-bearing sediment is crucial for understanding reservoir physical properties, yet traditional experimental methods are limited by insufficient resolution and inability to quantify cross-scale correlations between hydrate micro-morphology and macro-properties. This review comprehensively summarizes the latest advances in digital core technology and deep learning for hydrate-bearing sediment research, focusing on deep learning-driven techniques for image denoising, segmentation, enhancement, and 3D reconstruction, alongside physical property prediction. Therefore, deep learning techniques provide a revolutionary technological pathway for resolving the microstructural evolution of hydrate-bearing sediment and their macroscopic physical properties through intelligent image processing. We systematically evaluate traditional and deep learning-based methods through comparative case studies across diverse sediment types. Deep learning algorithms significantly outperform traditional filters in handling complex noise and segmenting low-contrast multiphase interfaces, while generative adversarial networks enable efficient 3D reconstruction from limited 2D data. However, deep learning-based prediction of macro-properties remains nascent due to challenges in data scarcity and model generalization. This review provides the first holistic synthesis of deep learning applications in hydrate-bearing sediment digital core analysis, highlighting its transformative potential in bridging the critical micro–macro gap—a key frontier for future hydrate resource assessment and exploitation.